deep-research Skill for Claude: What It Does and How to Install It

Makes Claude run a multi-phase research process and write a cited report, with an evidence log and HTML and PDF versions.

Installs: 10.3K; on skills.sh
Stars: 1.2K; on GitHub
Updated: Apr 2026; last commit

This skill makes Claude run a staged research pipeline and deliver a report with citations. You pick one of 4 modes: quick (3 phases), standard (6, the default), deep (8) or ultradeep (8 or more). It runs many searches in parallel, cross-checks sources, drafts, critiques and packages the report.

Output goes to ~/Documents/[Topic]_Research_[YYYYMMDD]/ as markdown, source, evidence and claim logs, a run manifest, HTML and PDF. It is not for simple lookups or quick questions. It runs Python scripts and makes many web requests, so read the safety notes.

Good for

Analysts who need a cited report, not a quick answer
Teams comparing technologies or reviewing the state of the art
Anyone who wants a source and claim trail they can check

Install deep-research

Claude Code

Copy this prompt.

Install the deep-research skill in Claude Code. Run this command:

git clone https://github.com/199-biotechnologies/claude-deep-research-skill.git ~/.claude/skills/deep-research

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Requires git.

Pick one route. The git clone puts the whole repo in the skills folder, as the README says. The skill runs Python scripts that need Python 3.9 or later. PDF output also needs weasyprint, and the README lists an optional search-cli tool for multi-provider search.

From the project README. Source.

What's in the skill

When it loads

Use when the user needs multi-source research with citation tracking, evidence persistence, and structured report generation. Triggers on "deep research", "comprehensive analysis", "research report", "compare X vs Y", "analyze trends", or "state of the art". Not for simple lookups, debugging, or questions answerable with 1-2 searches.

Sections in SKILL.md

Core Purpose
Decision Tree
Workflow Overview
Execution
Output Contract
When to Use / NOT Use

Key facts

Author: 199 Biotechnologies (repo owner)
License: MIT (README)
Installs on skills.sh: 10,262 (Nov 2, 2026)
GitHub stars: 1,204 for the whole 199-biotechnologies/claude-deep-research-skill repo (Nov 2, 2026)
Last repo commit: Apr 11, 2026
What it includes: 32 files in the repo including SKILL.md: 6 reference files, 2 templates, 4 JSON schemas, 10 Python scripts, 6 test files and requirements.txt
Needs: Python 3.9 or later (standard library only for the core, per requirements.txt). PDF output needs weasyprint. search-cli with at least one provider key is optional
Works in: Claude Code, per the README title and install path

What it includes

Quick, standard, deep and ultradeep
Stops for simple lookups and debugging. Standard is the default, and the modes differ in phases and time (2 to 5 up to 20 to 45 minutes).
Try: “Deep research in standard mode: compare PostgreSQL and Supabase for a small SaaS.”
Read · Decision tree and modes
Scope, plan, retrieve, triangulate, outline, synthesize, critique, refine, package
Retrieval runs 5 to 10 parallel searches and 2 to 3 sub-agents. Deep modes add a critique and refine loop.
Write · Phases 1 to 8
search-cli first, WebSearch as fallback, Exa MCP optional
methodology.md tells Claude to use search-cli for all searches if installed, which needs provider API keys, and to fall back to WebSearch.
Runs code · Search tools
sources.jsonl, evidence.jsonl, claims.jsonl and run_manifest.json
Scripts such as evidence_store.py and citation_manager.py append quotes, sources and claims to files in the output folder.
Write · Evidence store
Checks the report structure
Run on the finished report before delivery.
Runs code · validate_report.py
Checks DOIs and URLs, flags hallucination patterns
Sends requests to doi.org and HEAD requests to cited URLs.
Runs code · verify_citations.py
McKinsey-style HTML, then PDF with weasyprint
html-generation.md opens the HTML and PDF with the open command after they are built.
Runs code · md_to_html.py and PDF
Auto-continues reports over 18K words with sub-agents
continuation.md spawns further agents with context handed over.
Runs code · Long-report continuation

From the official docs

Permissions

Medium risk

Claude runs bundled Python scripts, searches the web and writes report files under your Documents folder. Do not put confidential details in the research question.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

Searches send your topic to search services
The methodology uses WebSearch, and search-cli if installed, which calls providers such as Brave, Serper, Exa, Jina and Firecrawl with your keys. Keep private details out of the question. github.com
Citation checks contact the sites you cite
verify_citations.py sends requests to doi.org and HEAD requests to each cited URL. The sites see the request from your machine. github.com
Optional installs: search-cli and weasyprint
The README suggests brew tap 199-biotechnologies/tap and brew install search-cli, and requirements.txt lists pip install weasyprint for PDFs. Review them before installing. github.com
Long reports spawn more agents
continuation.md spawns further agents for reports over 18K words, which uses more of your usage allowance. github.com

Security advisories

No published security advisories as of Nov 2, 2026.

Technical details

Turning on read-only	No read-only mode. Use version control.
Talks to	doi.org and the URLs you cite (verify_citations.py). Web search providers: WebSearch, and Brave, Serper, Exa, Jina or Firecrawl via the optional search-cli
Sign-in and keys	SKILL.md names no key for the core. search-cli keys are set with the search config command and stored by that tool.
Scopes	Claude runs Python scripts in the repo (validators, citation checks, HTML conversion) Output files go to ~/Documents/[Topic]_Research_[YYYYMMDD]/ html-generation.md uses the open command to open the HTML and PDF Phase 3 runs many web searches and spawns sub-agents

Questions

What does the deep-research skill do?

It makes Claude research a topic in stages and write a cited report. The skill keeps a log of sources, evidence and claims so you can check what backs each statement.

How do I install deep-research?

Use the install box at the top of this page. The README's own route is a git clone into the Claude Code skills folder. The repo is the skill, so the clone puts every file there.

Does it need an API key?

Not for basic use, per the README. The optional search-cli tool needs at least one search provider key (Brave, Serper, Exa, Jina or Firecrawl). Without it, SKILL.md's methodology falls back to WebSearch.

Does it work in the Claude app?

The install box has no Claude app upload for it, because the skill runs Python scripts, writes files to your Documents folder and opens them. Use Claude Code.

How long does a run take?

SKILL.md gives 2 to 5 minutes for quick, 5 to 10 for standard, 10 to 20 for deep and 20 to 45 for ultradeep.

When should I not use it?

SKILL.md says not to use it for simple lookups, debugging or questions answerable with one or two searches.

Is it safe?

It runs bundled Python scripts, makes many web requests and writes a report folder under Documents. Search queries and cited URLs leave your machine. If you install search-cli, your provider keys are stored by that tool. Run it on topics you are comfortable searching for.

Information

updated Nov 2, 2026

Repository: 199-biotechnologies/claude-deep-research-skill

199-biotechnologies/claude-deep-research-skill skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 6

deep-research SKILL.md · github.com
claude-deep-research-skill README · github.com
requirements.txt · github.com
methodology.md · github.com
html-generation.md · github.com
verify_citations.py · github.com